(Approx. 5 minutes de lecture)
(La suite de l’article en anglais) Cet article est le premier d’une nouvelle série de contenu éducatif de Maxa AI qui exploite l’intersection entre la performance d’affaires pour enterprises et l’intelligence machine. ///////
In today’s edition, Dennis Horton – Maxa AI’s Head of Data Science – will talk about business and financial forecasting and how to think about the massive headache that is COVID-19! We will introduce a few must-know data science concepts that underlie your business’ financial data.
At this point in time, the thought of eating at a crowded restaurant or sipping a beer in a bustling bar seems extremely foreign. It’s interesting how quickly we adapt to new living situations when things change. As we celebrated New Year’s eve 2019, we did not expect the 2020 we were soon to be handed.
Let’s rewind a few months ago: businesses everywhere were wrapping up 2019 and many were touching-up their 2020 budgets. Forecasts were being made based on managers’ experience and business intuition, the finance team’s structured support and insights, and common trade tools and techniques. FP&A folks (financial planning & analysis) are true corporate heroes. A ton of hard work – that now seems like a long time ago in a galaxy far, far away.
For most managers, the last few weeks have been about business continuity and cost control. Now that painful initiatives have been rolled out, and that sections of national economies are cautiously being reopened, executives are standing back up, looking to the horizon, shifting from a cost mindset to a revenue mindset, and wondering: so, how much longer will this last? What does recovery look like? Will it get back to normal? Or is there a new normal?
To look at possible scenarios for the remainder of 2020, let’s introduce a few basic data science concepts. Firstly, company financial data is typically logged as transactions and can be aggregated into “time series” – consistently defined data items at regular time points (often monthly) over time. An example time series could be: Acme Corp’s monthly sales of travel luggage for the last six years. Times series analysis is a powerful tool for financial forecasting. A diversified global company may have thousands (or millions!) of time series of business-financial data.
Any time series can be decomposed into three main components, namely:
Time Series = Trend + Seasonal + Irregular
The TREND represents the underlying long term behaviour of the series; it shouldn’t be influenced by one-off blips in activity but should focus on the long game. The SEASONAL component covers anything that is systematic and repeating with roughly the same magnitude each time period, often driven by seasons. The IRREGULAR component is anything that is left over after these two components have been estimated; the “leftovers” have two forms – random variations in that data (called “noise”) or extreme events that were not expected given the history of the series… like COVID-19!
Graphically, a time series of Acme’s luggage sales, with all three components, could look like this:
In the above graph, we can see a gradual, upward trend in luggage sales; there is a strong yearly seasonal pattern (higher luggage sales in the summer months), with random movements around this seasonal pattern (the irregular component).
Decomposing this series into its individual components is a nice way to visualise the many commercial dynamics at play. As an analogy, think of yourself listening to your favourite rock & roll hit, and then, for a few brief moments, trying to focus on just the bass guitar, or the drums, or the lyrics – screening everything else out. The following plot first shows the observed time series and then each of the components are plotted individually.
To better understand COVID-19’s impact on a business, keep in mind these three components above. If you remember only one thing from this article, remember this trilogy! Trend, seasonal, irregular.
Let’s go back to our example – Acme’s luggage sales – and consider four scenarios that can play out in Acme’s time series from March 2020, when its business was hit hard. Each scenario is exaggerated and simplified for sake of clarity. In real life, all four scenarios can happen simultaneously and to varying extents. This gets very complicated, very quickly. For now, let’s focus on our four simplified scenarios:
1) An extreme outlier: Let’s consider the unlikely scenario that a cheap, existing drug is proven effective against COVID-19, quickly distributed to everyone on the planet, and business swiftly returns to normal levels by May 2020. COVID-19 would be an example of an extreme outlier. We would see one blip in our series – specifically in the irregular component. The trend would not be affected since this is not a long term impact. An extreme outlier would look like this:
2) A transitional change: let’s now assume that an effective vaccine is fast-tracked and widely distributed, and that global business gets back to normal by August 2020. We call this a transitional change which is effectively several extreme outliers in a row. After this small group of extreme outliers, the series returns to the previous level and business continues as usual; the long term trend is not affected.
3) A trend break: For many business activities, the impact of COVID-19 will likely reach far into the future. The previous level of activity may never be experienced again, or at least not for a long period of time. This is a trend break as the level of the series has sustainably changed.
4) A seasonal break: The seasonal component can change gradually over time as social norms adapt and commercial behaviours change (for clarity, a “season” refers to a cyclical pattern independent of the meteorological seasons that the word commonly implies). However, seasonal breaks can occur when the yearly timing or the magnitude of activity changes suddenly and persistently. COVID-19 may have an impact on the seasonal activity for years to come as businesses and consumers shift, relax or double down on certain habits, especially when considering a diverse, global customer base.
These four scenarios illustrate the pandemic’s potential impacts on Acme’s luggage sales. As mentioned above, real life is complex – multiple impacts can combine into one hazy-looking scenario.
To augment our example with a real life twist, consider that we focused our discussion on a single Acme product (or one SKU#, in business jargon). Acme is a successful enterprise with 2,000 products and 10,000 customers, in 25 countries. How does this play out if every product (or product family) has a unique demand and revenue signature across many locations?
This is where automation and machine intelligence become key. The ability to detect and estimate – precisely, rapidly and at scale – multiple effects in time series is powerful to successfully weather the forecast storm induced by COVID-19. Monitoring individual time series within a business simultaneously aids in detecting early signs, discrepancies and recoveries, and allows adjustments to forecasts to be conducted on the fly.
By combining machine learning, software automation and FP&A best practices, one can provide a business with the best possible predictions for 2020 and beyond. No more “flying blind”!
Stay tuned for our next instalments as we dive into some of these challenges.